pytorch 官网教程

• 1.tensors 讲解了tensors相关操作

  • 生成张量
  • 从数组转换到张量

• 2.datasets & dataloaders ：PyTorch provides two data primitives:

  • torch.utils.data.DataLoader：wraps an iterable around the Dataset to enable easy access to the samples
  • torch.utils.data.Dataset：stores the samples and their corresponding labels
  • Loading a Dataset
    • root is the path where the train/test data is stored,
    • train specifies training or test dataset,
    • download=Truedownloads the data from the internet if it’s not available at root.
    • transform and target_transform specify the feature and label transformations

• 3.Transforms

  • transforms 是pytorch 提供的用来操作数据的模块
    • transform：to modify the features
    • target_transform：to modify the labels
    • from torchvision.transforms import ToTensor, Lambda
      • ToTensor()
        • ToTensor converts a PIL image or NumPy ndarray into a FloatTensor. and scales the image’s pixel intensity values in the range [0., 1.]
      • Lambda()
        • Lambda transforms apply any user-defined lambda function.

• 4.build models

  • import lib
  • Get Device for Training
  • Define the Model Class
  • Model Layers
  • Model Parameters

• 5.differentiation engine

• When training neural networks, the most frequently used algorithm is back propagation where parameters are adjusted according to the gradient of the loss function with respect to the given parameter.To compute those gradients, PyTorch provides torch.autograd.
• 6.Optimizing Model Parameters
  • Hyperparameters：需要人为调节的参数称为超参数
    • Number of Epochs:the number times to iterate over the dataset
      • Training a model is an iterative process; In each iteration (called an epoch)
        • guess the output
        • calculates loss
        • collects the derivatives of the error (with respect to its parameters)
        • optimizes parameters*(using gradient descent)
    • Batch Size:the number of data samples propagated through the network before the parameters are updated
    • Learning Rate how much to update models parameters at each batch/epoch. Smaller values yield slow learning speed
  • Optimization Loop:Each epoch consists of two main parts:
    • The Train Loop ： iterate over the training dataset and try to converge(收敛) to optimal parameters.
    • The Validation/Test Loop : iterate over the test dataset to check if model performance is improving.
  • Loss Function：三种常用的损失函数
    • nn.MSELoss (Mean Square Error)：for regression tasks
    • nn.NLLLoss (Negative Log Likelihood) for classification.
    • nn.CrossEntropyLosscombines nn.LogSoftmax and nn.NLLLoss.
  • Optimizer
    • optimizer = torch.optim.SGD(model.parameters(), lr=learning_rate)
• Save & Load Model